From 4bf8d1b20dfab8d4b07c837b4e5bbd34cd8cab41 Mon Sep 17 00:00:00 2001
From: nate <nate@forehead-gate.com>
Date: Thu, 9 Apr 2026 03:44:57 +0400
Subject: [PATCH] fix: improve state management

---
 apps/api/src/routes/internal.ts | 53 +++++++++++++++++++--------------
 1 file changed, 30 insertions(+), 23 deletions(-)

diff --git a/apps/api/src/routes/internal.ts b/apps/api/src/routes/internal.ts
index 17d55c4..c6f13b2 100644
--- a/apps/api/src/routes/internal.ts
+++ b/apps/api/src/routes/internal.ts
@@ -29,35 +29,42 @@ export const internal = new Elysia({ prefix: "/internal", detail: { hide: true }
 
   .get("/due", async ({ request }) => {
     const params = new URL(request.url).searchParams;
-    const region = params.get('region') || undefined;
+    const region = params.get('region') || 'default';
     const lookaheadMs = Math.min(Number(params.get('lookahead_ms') || 2000), 10000);
 
+    // No JOIN, no state lookup. The check schedule for a monitor is purely
+    // a function of its created_at and interval_s — every interval since
+    // creation is a tick. We pull all enabled monitors that match this
+    // region, compute the next tick in JS, and return the ones whose next
+    // tick falls within the lookahead window.
     const monitors = await sql`
-      SELECT m.id, m.url, m.method, m.request_headers, m.request_body, m.timeout_ms, m.interval_s, m.query, m.regions,
-             m.max_retries, m.retry_interval_s,
-             EXTRACT(EPOCH FROM (
-               CASE
-                 WHEN last.checked_at IS NULL THEN now()
-                 ELSE last.checked_at + (m.interval_s || ' seconds')::interval
-               END
-             ))::bigint * 1000 AS scheduled_at_ms
-      FROM monitors m
-      LEFT JOIN LATERAL (
-        SELECT checked_at FROM pings
-        WHERE monitor_id = m.id
-          AND (${region ? sql`region = ${region}` : sql`true`})
-        ORDER BY checked_at DESC LIMIT 1
-      ) last ON true
-      WHERE m.enabled = true
-        AND (last.checked_at IS NULL
-          OR last.checked_at < now() - (m.interval_s || ' seconds')::interval + (${lookaheadMs} || ' milliseconds')::interval)
+      SELECT id, url, method, request_headers, request_body, timeout_ms, interval_s, query, regions,
+             max_retries, retry_interval_s, created_at
+      FROM monitors
+      WHERE enabled = true
         AND (
-          array_length(m.regions, 1) IS NULL
-          OR m.regions = '{}'
-          OR ${region ? sql`${region} = ANY(m.regions)` : sql`true`}
+          array_length(regions, 1) IS NULL
+          OR regions = '{}'
+          OR ${region} = ANY(regions)
         )
+      LIMIT 500
     `;
-    return monitors;
+
+    const nowMs = Date.now();
+    const lookaheadEnd = nowMs + lookaheadMs;
+    const due: any[] = [];
+    for (const m of monitors) {
+      const createdMs = new Date((m as any).created_at).getTime();
+      const intervalMs = Number((m as any).interval_s) * 1000;
+      if (intervalMs <= 0) continue;
+      const elapsed = nowMs - createdMs;
+      const ticksPassed = Math.floor(elapsed / intervalMs);
+      const nextTickMs = createdMs + (ticksPassed + 1) * intervalMs;
+      if (nextTickMs <= lookaheadEnd) {
+        due.push({ ...m, scheduled_at_ms: nextTickMs });
+      }
+    }
+    return due;
   })
 
   .post("/prune", async () => {